blob: 4a944676358e841821a4b0d9a53ff312d2de9521 [file] [log] [blame]
Don Zickus58687ac2010-05-07 17:11:44 -04001/*
2 * Detect hard and soft lockups on a system
3 *
4 * started by Don Zickus, Copyright (C) 2010 Red Hat, Inc.
5 *
Fernando Luis Vázquez Cao86f5e6a2012-02-09 17:42:22 -05006 * Note: Most of this code is borrowed heavily from the original softlockup
7 * detector, so thanks to Ingo for the initial implementation.
8 * Some chunks also taken from the old x86-specific nmi watchdog code, thanks
Don Zickus58687ac2010-05-07 17:11:44 -04009 * to those contributors as well.
10 */
11
Andrew Morton45019802012-03-23 15:01:55 -070012#define pr_fmt(fmt) "NMI watchdog: " fmt
13
Don Zickus58687ac2010-05-07 17:11:44 -040014#include <linux/mm.h>
15#include <linux/cpu.h>
16#include <linux/nmi.h>
17#include <linux/init.h>
18#include <linux/delay.h>
19#include <linux/freezer.h>
20#include <linux/kthread.h>
21#include <linux/lockdep.h>
22#include <linux/notifier.h>
23#include <linux/module.h>
24#include <linux/sysctl.h>
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +000025#include <linux/smpboot.h>
Clark Williams8bd75c72013-02-07 09:47:07 -060026#include <linux/sched/rt.h>
Don Zickus58687ac2010-05-07 17:11:44 -040027
28#include <asm/irq_regs.h>
Eric B Munson5d1c0f42012-03-10 14:37:28 -050029#include <linux/kvm_para.h>
Don Zickus58687ac2010-05-07 17:11:44 -040030#include <linux/perf_event.h>
31
Marcin Slusarz41350382011-01-28 11:00:31 -050032int watchdog_enabled = 1;
Mandeep Singh Baines4eec42f2011-05-22 22:10:23 -070033int __read_mostly watchdog_thresh = 10;
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +000034static int __read_mostly watchdog_disabled;
Chuansheng Liu0f34c402012-12-17 15:59:50 -080035static u64 __read_mostly sample_period;
Don Zickus58687ac2010-05-07 17:11:44 -040036
37static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts);
38static DEFINE_PER_CPU(struct task_struct *, softlockup_watchdog);
39static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer);
40static DEFINE_PER_CPU(bool, softlockup_touch_sync);
Don Zickus58687ac2010-05-07 17:11:44 -040041static DEFINE_PER_CPU(bool, soft_watchdog_warn);
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +000042static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts);
43static DEFINE_PER_CPU(unsigned long, soft_lockup_hrtimer_cnt);
Frederic Weisbecker23637d42010-05-15 23:15:20 +020044#ifdef CONFIG_HARDLOCKUP_DETECTOR
Don Zickuscafcd802010-05-14 11:11:21 -040045static DEFINE_PER_CPU(bool, hard_watchdog_warn);
46static DEFINE_PER_CPU(bool, watchdog_nmi_touch);
Don Zickus58687ac2010-05-07 17:11:44 -040047static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved);
48static DEFINE_PER_CPU(struct perf_event *, watchdog_ev);
49#endif
50
Don Zickus58687ac2010-05-07 17:11:44 -040051/* boot commands */
52/*
53 * Should we panic when a soft-lockup or hard-lockup occurs:
54 */
Frederic Weisbecker23637d42010-05-15 23:15:20 +020055#ifdef CONFIG_HARDLOCKUP_DETECTOR
Don Zickusfef2c9b2011-03-22 16:34:16 -070056static int hardlockup_panic =
57 CONFIG_BOOTPARAM_HARDLOCKUP_PANIC_VALUE;
Don Zickus58687ac2010-05-07 17:11:44 -040058
59static int __init hardlockup_panic_setup(char *str)
60{
61 if (!strncmp(str, "panic", 5))
62 hardlockup_panic = 1;
Don Zickusfef2c9b2011-03-22 16:34:16 -070063 else if (!strncmp(str, "nopanic", 7))
64 hardlockup_panic = 0;
Don Zickus5dc30552010-11-29 17:07:17 -050065 else if (!strncmp(str, "0", 1))
Marcin Slusarz41350382011-01-28 11:00:31 -050066 watchdog_enabled = 0;
Don Zickus58687ac2010-05-07 17:11:44 -040067 return 1;
68}
69__setup("nmi_watchdog=", hardlockup_panic_setup);
70#endif
71
72unsigned int __read_mostly softlockup_panic =
73 CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE;
74
75static int __init softlockup_panic_setup(char *str)
76{
77 softlockup_panic = simple_strtoul(str, NULL, 0);
78
79 return 1;
80}
81__setup("softlockup_panic=", softlockup_panic_setup);
82
83static int __init nowatchdog_setup(char *str)
84{
Marcin Slusarz41350382011-01-28 11:00:31 -050085 watchdog_enabled = 0;
Don Zickus58687ac2010-05-07 17:11:44 -040086 return 1;
87}
88__setup("nowatchdog", nowatchdog_setup);
89
90/* deprecated */
91static int __init nosoftlockup_setup(char *str)
92{
Marcin Slusarz41350382011-01-28 11:00:31 -050093 watchdog_enabled = 0;
Don Zickus58687ac2010-05-07 17:11:44 -040094 return 1;
95}
96__setup("nosoftlockup", nosoftlockup_setup);
97/* */
98
Mandeep Singh Baines4eec42f2011-05-22 22:10:23 -070099/*
100 * Hard-lockup warnings should be triggered after just a few seconds. Soft-
101 * lockups can have false positives under extreme conditions. So we generally
102 * want a higher threshold for soft lockups than for hard lockups. So we couple
103 * the thresholds with a factor: we make the soft threshold twice the amount of
104 * time the hard threshold is.
105 */
Ingo Molnar6e9101a2011-05-24 05:43:18 +0200106static int get_softlockup_thresh(void)
Mandeep Singh Baines4eec42f2011-05-22 22:10:23 -0700107{
108 return watchdog_thresh * 2;
109}
Don Zickus58687ac2010-05-07 17:11:44 -0400110
111/*
112 * Returns seconds, approximately. We don't need nanosecond
113 * resolution, and we don't need to waste time with a big divide when
114 * 2^30ns == 1.074s.
115 */
Namhyung Kimc06b4f12012-12-27 11:49:44 +0900116static unsigned long get_timestamp(void)
Don Zickus58687ac2010-05-07 17:11:44 -0400117{
Namhyung Kimc06b4f12012-12-27 11:49:44 +0900118 return local_clock() >> 30LL; /* 2^30 ~= 10^9 */
Don Zickus58687ac2010-05-07 17:11:44 -0400119}
120
Chuansheng Liu0f34c402012-12-17 15:59:50 -0800121static void set_sample_period(void)
Don Zickus58687ac2010-05-07 17:11:44 -0400122{
123 /*
Mandeep Singh Baines586692a2011-05-22 22:10:22 -0700124 * convert watchdog_thresh from seconds to ns
Fernando Luis Vázquez Cao86f5e6a2012-02-09 17:42:22 -0500125 * the divide by 5 is to give hrtimer several chances (two
126 * or three with the current relation between the soft
127 * and hard thresholds) to increment before the
128 * hardlockup detector generates a warning
Don Zickus58687ac2010-05-07 17:11:44 -0400129 */
Chuansheng Liu0f34c402012-12-17 15:59:50 -0800130 sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5);
Don Zickus58687ac2010-05-07 17:11:44 -0400131}
132
133/* Commands for resetting the watchdog */
134static void __touch_watchdog(void)
135{
Namhyung Kimc06b4f12012-12-27 11:49:44 +0900136 __this_cpu_write(watchdog_touch_ts, get_timestamp());
Don Zickus58687ac2010-05-07 17:11:44 -0400137}
138
Don Zickus332fbdb2010-05-07 17:11:45 -0400139void touch_softlockup_watchdog(void)
Don Zickus58687ac2010-05-07 17:11:44 -0400140{
Christoph Lameter909ea962010-12-08 16:22:55 +0100141 __this_cpu_write(watchdog_touch_ts, 0);
Don Zickus58687ac2010-05-07 17:11:44 -0400142}
Ingo Molnar0167c782010-05-13 08:53:33 +0200143EXPORT_SYMBOL(touch_softlockup_watchdog);
Don Zickus58687ac2010-05-07 17:11:44 -0400144
Don Zickus332fbdb2010-05-07 17:11:45 -0400145void touch_all_softlockup_watchdogs(void)
Don Zickus58687ac2010-05-07 17:11:44 -0400146{
147 int cpu;
148
149 /*
150 * this is done lockless
151 * do we care if a 0 races with a timestamp?
152 * all it means is the softlock check starts one cycle later
153 */
154 for_each_online_cpu(cpu)
155 per_cpu(watchdog_touch_ts, cpu) = 0;
156}
157
Don Zickuscafcd802010-05-14 11:11:21 -0400158#ifdef CONFIG_HARDLOCKUP_DETECTOR
Don Zickus58687ac2010-05-07 17:11:44 -0400159void touch_nmi_watchdog(void)
160{
Don Zickus68d3f1d2010-08-31 23:00:07 -0400161 if (watchdog_enabled) {
162 unsigned cpu;
163
164 for_each_present_cpu(cpu) {
165 if (per_cpu(watchdog_nmi_touch, cpu) != true)
166 per_cpu(watchdog_nmi_touch, cpu) = true;
167 }
168 }
Don Zickus332fbdb2010-05-07 17:11:45 -0400169 touch_softlockup_watchdog();
Don Zickus58687ac2010-05-07 17:11:44 -0400170}
171EXPORT_SYMBOL(touch_nmi_watchdog);
172
Don Zickuscafcd802010-05-14 11:11:21 -0400173#endif
174
Don Zickus58687ac2010-05-07 17:11:44 -0400175void touch_softlockup_watchdog_sync(void)
176{
177 __raw_get_cpu_var(softlockup_touch_sync) = true;
178 __raw_get_cpu_var(watchdog_touch_ts) = 0;
179}
180
Frederic Weisbecker23637d42010-05-15 23:15:20 +0200181#ifdef CONFIG_HARDLOCKUP_DETECTOR
Don Zickus58687ac2010-05-07 17:11:44 -0400182/* watchdog detector functions */
Don Zickus26e09c62010-05-17 18:06:04 -0400183static int is_hardlockup(void)
Don Zickus58687ac2010-05-07 17:11:44 -0400184{
Christoph Lameter909ea962010-12-08 16:22:55 +0100185 unsigned long hrint = __this_cpu_read(hrtimer_interrupts);
Don Zickus58687ac2010-05-07 17:11:44 -0400186
Christoph Lameter909ea962010-12-08 16:22:55 +0100187 if (__this_cpu_read(hrtimer_interrupts_saved) == hrint)
Don Zickus58687ac2010-05-07 17:11:44 -0400188 return 1;
189
Christoph Lameter909ea962010-12-08 16:22:55 +0100190 __this_cpu_write(hrtimer_interrupts_saved, hrint);
Don Zickus58687ac2010-05-07 17:11:44 -0400191 return 0;
192}
193#endif
194
Don Zickus26e09c62010-05-17 18:06:04 -0400195static int is_softlockup(unsigned long touch_ts)
Don Zickus58687ac2010-05-07 17:11:44 -0400196{
Namhyung Kimc06b4f12012-12-27 11:49:44 +0900197 unsigned long now = get_timestamp();
Don Zickus58687ac2010-05-07 17:11:44 -0400198
199 /* Warn about unreasonable delays: */
Mandeep Singh Baines4eec42f2011-05-22 22:10:23 -0700200 if (time_after(now, touch_ts + get_softlockup_thresh()))
Don Zickus58687ac2010-05-07 17:11:44 -0400201 return now - touch_ts;
202
203 return 0;
204}
205
Frederic Weisbecker23637d42010-05-15 23:15:20 +0200206#ifdef CONFIG_HARDLOCKUP_DETECTOR
Cyrill Gorcunov1880c4a2011-06-23 16:49:18 +0400207
Don Zickus58687ac2010-05-07 17:11:44 -0400208static struct perf_event_attr wd_hw_attr = {
209 .type = PERF_TYPE_HARDWARE,
210 .config = PERF_COUNT_HW_CPU_CYCLES,
211 .size = sizeof(struct perf_event_attr),
212 .pinned = 1,
213 .disabled = 1,
214};
215
216/* Callback function for perf event subsystem */
Peter Zijlstraa8b0ca12011-06-27 14:41:57 +0200217static void watchdog_overflow_callback(struct perf_event *event,
Don Zickus58687ac2010-05-07 17:11:44 -0400218 struct perf_sample_data *data,
219 struct pt_regs *regs)
220{
Peter Zijlstrac6db67c2010-08-20 11:49:15 +0200221 /* Ensure the watchdog never gets throttled */
222 event->hw.interrupts = 0;
223
Christoph Lameter909ea962010-12-08 16:22:55 +0100224 if (__this_cpu_read(watchdog_nmi_touch) == true) {
225 __this_cpu_write(watchdog_nmi_touch, false);
Don Zickus58687ac2010-05-07 17:11:44 -0400226 return;
227 }
228
229 /* check for a hardlockup
230 * This is done by making sure our timer interrupt
231 * is incrementing. The timer interrupt should have
232 * fired multiple times before we overflow'd. If it hasn't
233 * then this is a good indication the cpu is stuck
234 */
Don Zickus26e09c62010-05-17 18:06:04 -0400235 if (is_hardlockup()) {
236 int this_cpu = smp_processor_id();
237
Don Zickus58687ac2010-05-07 17:11:44 -0400238 /* only print hardlockups once */
Christoph Lameter909ea962010-12-08 16:22:55 +0100239 if (__this_cpu_read(hard_watchdog_warn) == true)
Don Zickus58687ac2010-05-07 17:11:44 -0400240 return;
241
242 if (hardlockup_panic)
243 panic("Watchdog detected hard LOCKUP on cpu %d", this_cpu);
244 else
245 WARN(1, "Watchdog detected hard LOCKUP on cpu %d", this_cpu);
246
Christoph Lameter909ea962010-12-08 16:22:55 +0100247 __this_cpu_write(hard_watchdog_warn, true);
Don Zickus58687ac2010-05-07 17:11:44 -0400248 return;
249 }
250
Christoph Lameter909ea962010-12-08 16:22:55 +0100251 __this_cpu_write(hard_watchdog_warn, false);
Don Zickus58687ac2010-05-07 17:11:44 -0400252 return;
253}
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000254#endif /* CONFIG_HARDLOCKUP_DETECTOR */
255
Don Zickus58687ac2010-05-07 17:11:44 -0400256static void watchdog_interrupt_count(void)
257{
Christoph Lameter909ea962010-12-08 16:22:55 +0100258 __this_cpu_inc(hrtimer_interrupts);
Don Zickus58687ac2010-05-07 17:11:44 -0400259}
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000260
261static int watchdog_nmi_enable(unsigned int cpu);
262static void watchdog_nmi_disable(unsigned int cpu);
Don Zickus58687ac2010-05-07 17:11:44 -0400263
264/* watchdog kicker functions */
265static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer)
266{
Christoph Lameter909ea962010-12-08 16:22:55 +0100267 unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts);
Don Zickus58687ac2010-05-07 17:11:44 -0400268 struct pt_regs *regs = get_irq_regs();
269 int duration;
270
271 /* kick the hardlockup detector */
272 watchdog_interrupt_count();
273
274 /* kick the softlockup detector */
Christoph Lameter909ea962010-12-08 16:22:55 +0100275 wake_up_process(__this_cpu_read(softlockup_watchdog));
Don Zickus58687ac2010-05-07 17:11:44 -0400276
277 /* .. and repeat */
Chuansheng Liu0f34c402012-12-17 15:59:50 -0800278 hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period));
Don Zickus58687ac2010-05-07 17:11:44 -0400279
280 if (touch_ts == 0) {
Christoph Lameter909ea962010-12-08 16:22:55 +0100281 if (unlikely(__this_cpu_read(softlockup_touch_sync))) {
Don Zickus58687ac2010-05-07 17:11:44 -0400282 /*
283 * If the time stamp was touched atomically
284 * make sure the scheduler tick is up to date.
285 */
Christoph Lameter909ea962010-12-08 16:22:55 +0100286 __this_cpu_write(softlockup_touch_sync, false);
Don Zickus58687ac2010-05-07 17:11:44 -0400287 sched_clock_tick();
288 }
Eric B Munson5d1c0f42012-03-10 14:37:28 -0500289
290 /* Clear the guest paused flag on watchdog reset */
291 kvm_check_and_clear_guest_paused();
Don Zickus58687ac2010-05-07 17:11:44 -0400292 __touch_watchdog();
293 return HRTIMER_RESTART;
294 }
295
296 /* check for a softlockup
297 * This is done by making sure a high priority task is
298 * being scheduled. The task touches the watchdog to
299 * indicate it is getting cpu time. If it hasn't then
300 * this is a good indication some task is hogging the cpu
301 */
Don Zickus26e09c62010-05-17 18:06:04 -0400302 duration = is_softlockup(touch_ts);
Don Zickus58687ac2010-05-07 17:11:44 -0400303 if (unlikely(duration)) {
Eric B Munson5d1c0f42012-03-10 14:37:28 -0500304 /*
305 * If a virtual machine is stopped by the host it can look to
306 * the watchdog like a soft lockup, check to see if the host
307 * stopped the vm before we issue the warning
308 */
309 if (kvm_check_and_clear_guest_paused())
310 return HRTIMER_RESTART;
311
Don Zickus58687ac2010-05-07 17:11:44 -0400312 /* only warn once */
Christoph Lameter909ea962010-12-08 16:22:55 +0100313 if (__this_cpu_read(soft_watchdog_warn) == true)
Don Zickus58687ac2010-05-07 17:11:44 -0400314 return HRTIMER_RESTART;
315
Prarit Bhargavab0f4c4b2012-01-26 08:55:34 -0500316 printk(KERN_EMERG "BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n",
Don Zickus26e09c62010-05-17 18:06:04 -0400317 smp_processor_id(), duration,
Don Zickus58687ac2010-05-07 17:11:44 -0400318 current->comm, task_pid_nr(current));
319 print_modules();
320 print_irqtrace_events(current);
321 if (regs)
322 show_regs(regs);
323 else
324 dump_stack();
325
326 if (softlockup_panic)
327 panic("softlockup: hung tasks");
Christoph Lameter909ea962010-12-08 16:22:55 +0100328 __this_cpu_write(soft_watchdog_warn, true);
Don Zickus58687ac2010-05-07 17:11:44 -0400329 } else
Christoph Lameter909ea962010-12-08 16:22:55 +0100330 __this_cpu_write(soft_watchdog_warn, false);
Don Zickus58687ac2010-05-07 17:11:44 -0400331
332 return HRTIMER_RESTART;
333}
334
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000335static void watchdog_set_prio(unsigned int policy, unsigned int prio)
Don Zickus58687ac2010-05-07 17:11:44 -0400336{
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000337 struct sched_param param = { .sched_priority = prio };
338
339 sched_setscheduler(current, policy, &param);
340}
341
342static void watchdog_enable(unsigned int cpu)
343{
Don Zickus26e09c62010-05-17 18:06:04 -0400344 struct hrtimer *hrtimer = &__raw_get_cpu_var(watchdog_hrtimer);
Don Zickus58687ac2010-05-07 17:11:44 -0400345
Bjørn Mork3935e8952012-12-19 20:51:31 +0100346 /* kick off the timer for the hardlockup detector */
347 hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
348 hrtimer->function = watchdog_timer_fn;
349
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000350 if (!watchdog_enabled) {
351 kthread_park(current);
352 return;
353 }
354
355 /* Enable the perf event */
356 watchdog_nmi_enable(cpu);
Don Zickus58687ac2010-05-07 17:11:44 -0400357
Don Zickus58687ac2010-05-07 17:11:44 -0400358 /* done here because hrtimer_start can only pin to smp_processor_id() */
Chuansheng Liu0f34c402012-12-17 15:59:50 -0800359 hrtimer_start(hrtimer, ns_to_ktime(sample_period),
Don Zickus58687ac2010-05-07 17:11:44 -0400360 HRTIMER_MODE_REL_PINNED);
361
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000362 /* initialize timestamp */
363 watchdog_set_prio(SCHED_FIFO, MAX_RT_PRIO - 1);
364 __touch_watchdog();
Don Zickus58687ac2010-05-07 17:11:44 -0400365}
366
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000367static void watchdog_disable(unsigned int cpu)
368{
369 struct hrtimer *hrtimer = &__raw_get_cpu_var(watchdog_hrtimer);
370
371 watchdog_set_prio(SCHED_NORMAL, 0);
372 hrtimer_cancel(hrtimer);
373 /* disable the perf event */
374 watchdog_nmi_disable(cpu);
375}
376
377static int watchdog_should_run(unsigned int cpu)
378{
379 return __this_cpu_read(hrtimer_interrupts) !=
380 __this_cpu_read(soft_lockup_hrtimer_cnt);
381}
382
383/*
384 * The watchdog thread function - touches the timestamp.
385 *
Chuansheng Liu0f34c402012-12-17 15:59:50 -0800386 * It only runs once every sample_period seconds (4 seconds by
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000387 * default) to reset the softlockup timestamp. If this gets delayed
388 * for more than 2*watchdog_thresh seconds then the debug-printout
389 * triggers in watchdog_timer_fn().
390 */
391static void watchdog(unsigned int cpu)
392{
393 __this_cpu_write(soft_lockup_hrtimer_cnt,
394 __this_cpu_read(hrtimer_interrupts));
395 __touch_watchdog();
396}
Don Zickus58687ac2010-05-07 17:11:44 -0400397
Frederic Weisbecker23637d42010-05-15 23:15:20 +0200398#ifdef CONFIG_HARDLOCKUP_DETECTOR
Don Zickusa7027042012-06-13 09:35:48 -0400399/*
400 * People like the simple clean cpu node info on boot.
401 * Reduce the watchdog noise by only printing messages
402 * that are different from what cpu0 displayed.
403 */
404static unsigned long cpu0_err;
405
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000406static int watchdog_nmi_enable(unsigned int cpu)
Don Zickus58687ac2010-05-07 17:11:44 -0400407{
408 struct perf_event_attr *wd_attr;
409 struct perf_event *event = per_cpu(watchdog_ev, cpu);
410
411 /* is it already setup and enabled? */
412 if (event && event->state > PERF_EVENT_STATE_OFF)
413 goto out;
414
415 /* it is setup but not enabled */
416 if (event != NULL)
417 goto out_enable;
418
Don Zickus58687ac2010-05-07 17:11:44 -0400419 wd_attr = &wd_hw_attr;
Mandeep Singh Baines4eec42f2011-05-22 22:10:23 -0700420 wd_attr->sample_period = hw_nmi_get_sample_period(watchdog_thresh);
Cyrill Gorcunov1880c4a2011-06-23 16:49:18 +0400421
422 /* Try to register using hardware perf events */
Avi Kivity4dc0da82011-06-29 18:42:35 +0300423 event = perf_event_create_kernel_counter(wd_attr, cpu, NULL, watchdog_overflow_callback, NULL);
Don Zickusa7027042012-06-13 09:35:48 -0400424
425 /* save cpu0 error for future comparision */
426 if (cpu == 0 && IS_ERR(event))
427 cpu0_err = PTR_ERR(event);
428
Don Zickus58687ac2010-05-07 17:11:44 -0400429 if (!IS_ERR(event)) {
Don Zickusa7027042012-06-13 09:35:48 -0400430 /* only print for cpu0 or different than cpu0 */
431 if (cpu == 0 || cpu0_err)
432 pr_info("enabled on all CPUs, permanently consumes one hw-PMU counter.\n");
Don Zickus58687ac2010-05-07 17:11:44 -0400433 goto out_save;
434 }
435
Don Zickusa7027042012-06-13 09:35:48 -0400436 /* skip displaying the same error again */
437 if (cpu > 0 && (PTR_ERR(event) == cpu0_err))
438 return PTR_ERR(event);
Don Zickus5651f7f2011-02-09 14:02:33 -0500439
440 /* vary the KERN level based on the returned errno */
441 if (PTR_ERR(event) == -EOPNOTSUPP)
Andrew Morton45019802012-03-23 15:01:55 -0700442 pr_info("disabled (cpu%i): not supported (no LAPIC?)\n", cpu);
Don Zickus5651f7f2011-02-09 14:02:33 -0500443 else if (PTR_ERR(event) == -ENOENT)
Andrew Morton45019802012-03-23 15:01:55 -0700444 pr_warning("disabled (cpu%i): hardware events not enabled\n",
445 cpu);
Don Zickus5651f7f2011-02-09 14:02:33 -0500446 else
Andrew Morton45019802012-03-23 15:01:55 -0700447 pr_err("disabled (cpu%i): unable to create perf event: %ld\n",
448 cpu, PTR_ERR(event));
Akinobu Mitaeac24332010-08-31 23:00:08 -0400449 return PTR_ERR(event);
Don Zickus58687ac2010-05-07 17:11:44 -0400450
451 /* success path */
452out_save:
453 per_cpu(watchdog_ev, cpu) = event;
454out_enable:
455 perf_event_enable(per_cpu(watchdog_ev, cpu));
456out:
457 return 0;
458}
459
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000460static void watchdog_nmi_disable(unsigned int cpu)
Don Zickus58687ac2010-05-07 17:11:44 -0400461{
462 struct perf_event *event = per_cpu(watchdog_ev, cpu);
463
464 if (event) {
465 perf_event_disable(event);
466 per_cpu(watchdog_ev, cpu) = NULL;
467
468 /* should be in cleanup, but blocks oprofile */
469 perf_event_release_kernel(event);
470 }
471 return;
472}
473#else
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000474static int watchdog_nmi_enable(unsigned int cpu) { return 0; }
475static void watchdog_nmi_disable(unsigned int cpu) { return; }
Frederic Weisbecker23637d42010-05-15 23:15:20 +0200476#endif /* CONFIG_HARDLOCKUP_DETECTOR */
Don Zickus58687ac2010-05-07 17:11:44 -0400477
478/* prepare/enable/disable routines */
Vasily Averin4ff81952011-10-31 17:11:18 -0700479/* sysctl functions */
480#ifdef CONFIG_SYSCTL
Don Zickus58687ac2010-05-07 17:11:44 -0400481static void watchdog_enable_all_cpus(void)
482{
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000483 unsigned int cpu;
Marcin Slusarz39735762011-01-28 11:00:32 -0500484
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000485 if (watchdog_disabled) {
486 watchdog_disabled = 0;
487 for_each_online_cpu(cpu)
488 kthread_unpark(per_cpu(softlockup_watchdog, cpu));
489 }
Don Zickus58687ac2010-05-07 17:11:44 -0400490}
491
492static void watchdog_disable_all_cpus(void)
493{
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000494 unsigned int cpu;
Don Zickus58687ac2010-05-07 17:11:44 -0400495
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000496 if (!watchdog_disabled) {
497 watchdog_disabled = 1;
498 for_each_online_cpu(cpu)
499 kthread_park(per_cpu(softlockup_watchdog, cpu));
500 }
Don Zickus58687ac2010-05-07 17:11:44 -0400501}
502
Don Zickus58687ac2010-05-07 17:11:44 -0400503/*
Mandeep Singh Baines586692a2011-05-22 22:10:22 -0700504 * proc handler for /proc/sys/kernel/nmi_watchdog,watchdog_thresh
Don Zickus58687ac2010-05-07 17:11:44 -0400505 */
506
Mandeep Singh Baines586692a2011-05-22 22:10:22 -0700507int proc_dowatchdog(struct ctl_table *table, int write,
508 void __user *buffer, size_t *lenp, loff_t *ppos)
Don Zickus58687ac2010-05-07 17:11:44 -0400509{
Mandeep Singh Bainese04ab2b2011-05-22 22:10:21 -0700510 int ret;
Don Zickus58687ac2010-05-07 17:11:44 -0400511
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000512 if (watchdog_disabled < 0)
513 return -ENODEV;
514
Mandeep Singh Baines586692a2011-05-22 22:10:22 -0700515 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
Mandeep Singh Bainese04ab2b2011-05-22 22:10:21 -0700516 if (ret || !write)
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000517 return ret;
Mandeep Singh Bainese04ab2b2011-05-22 22:10:21 -0700518
Chuansheng Liu0f34c402012-12-17 15:59:50 -0800519 set_sample_period();
Mandeep Singh Baines586692a2011-05-22 22:10:22 -0700520 if (watchdog_enabled && watchdog_thresh)
Mandeep Singh Bainese04ab2b2011-05-22 22:10:21 -0700521 watchdog_enable_all_cpus();
522 else
523 watchdog_disable_all_cpus();
524
Mandeep Singh Bainese04ab2b2011-05-22 22:10:21 -0700525 return ret;
Don Zickus58687ac2010-05-07 17:11:44 -0400526}
Don Zickus58687ac2010-05-07 17:11:44 -0400527#endif /* CONFIG_SYSCTL */
528
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000529static struct smp_hotplug_thread watchdog_threads = {
530 .store = &softlockup_watchdog,
531 .thread_should_run = watchdog_should_run,
532 .thread_fn = watchdog,
533 .thread_comm = "watchdog/%u",
534 .setup = watchdog_enable,
535 .park = watchdog_disable,
536 .unpark = watchdog_enable,
Don Zickus58687ac2010-05-07 17:11:44 -0400537};
538
Peter Zijlstra004417a2010-11-25 18:38:29 +0100539void __init lockup_detector_init(void)
Don Zickus58687ac2010-05-07 17:11:44 -0400540{
Chuansheng Liu0f34c402012-12-17 15:59:50 -0800541 set_sample_period();
Thomas Gleixnerbcd951c2012-07-16 10:42:38 +0000542 if (smpboot_register_percpu_thread(&watchdog_threads)) {
543 pr_err("Failed to create watchdog threads, disabled\n");
544 watchdog_disabled = -ENODEV;
545 }
Don Zickus58687ac2010-05-07 17:11:44 -0400546}